Broadcom Qumran Architecture

The Broadcom Qumran family is tailored for access and aggregation routers. While it shares DNX architecture roots with Jericho, Qumran ASICs are typically deployed in fixed-form platforms and edge routing roles (carrier Ethernet, 5G mobile backhaul, broadband access).

Deep Packet Buffering & VOQ

Standard data center ToR switches rely on shallow, fast on-chip memory (e.g., 16 MB or 32 MB) to handle clean, evenly distributed traffic. However, service provider edge networks frequently encounter massive speed mismatches—such as funneling traffic from a 100G core down to a 10G customer link or cell tower.

To prevent devastating buffer exhaustion ("tail drops") in these scenarios, Qumran architecture pairs a small integrated on-chip buffer with massive external DRAM modules (often ranging from 2 GB up to 8 GB).

When congestion occurs, the ASIC seamlessly shunts overflowing packets into the external deep buffer. Crucially, Qumran employs Virtual Output Queuing (VOQ). Rather than blindly pushing packets to an egress port buffer until it overflows, packets are queued logically at their ingress port until the egress port is ready to transmit them. This totally eliminates head-of-line blocking and ensures high priority traffic (like PTP or voice) is never dropped behind a burst of bulk data.

Qumran-MX vs. Qumran-AX

While both share the same deep buffer and feature-rich routing DNA, they target different segments of the network edge:

Platforms Using This Architecture

References

See Also

← Back to the main buffer table